quARtet Marker: A 3D-Printable Multi-Tag Fiducial for Robust Near-Frontal Pose Estimation
This study addresses the low pose estimation accuracy and frequent grasping conflicts encountered with transparent or reflective laboratory vessels under near-frontal viewpoints. To overcome these challenges, we propose a 3D-printed multi-fiducial marker benchmark that arranges four AprilTags obliquely within a single compact volume to recover perspective cues, innovatively balancing pose consistency with the contact surface requirements of parallel-jaw grippers. By integrating AprilTag detection with a Perspective-n-Point (PnP) solver, the proposed method reduces the mean orientation error from 2.18° to below 0.47° while significantly decreasing positional error. Furthermore, this work reveals the inherent trade-offs between estimation accuracy and graspability across different marker layouts, providing practical design guidelines for robotic manipulation in laboratory environments.